Tax-benefit microsimulation
PolicyEngine microsimulation
Checked rule by rule, one external costing benchmark, and the limits.
Headline numbers
Household maths is exact; population costings are survey-weighted estimates with no interval.
| Dimension | Level | What would change it |
|---|---|---|
| Implementation fidelity | strong | Measured per-country statutory test coverage |
| Predictive validation | not applicable | None; not applicable by construction |
| Identification | not applicable | None |
| Policy counterfactuals | moderate | A US reform benchmarked to an independent costing |
| Uncertainty | weak | Bootstrap ranges on population aggregates |
| Vintage reproducibility | moderate | Aggregates reproducible without gated microdata |
The two checks
Rule arithmetic by hand, and one costing against HMRC.
1
Checked rule by rule
No forecast error to report: the model applies published rules.
About this stepnotes
Exact rule arithmetic
- Check
- the worked £50,000 case, by pencil
- Where
- the Use it tab
- Verifies
- the implemented rules
- Does not verify
- that every area is covered by a test
A hand check verifies the implemented rules, not that every area of legislation is covered by a test.
2
The one external comparison
A survey-weighted costing against HMRC’s post-behavioural ready reckoner.
About this step6 points
1p on the basic rate
- Ours, 2026
- £6.55bn, static
- HMRC, 2026–27
- £6.9bn, post-behavioural
- Ours vs HMRC, 2028–29
- £7.01bn vs £8.2bn (−14.5%)
- Ours, 2030
- £7.48bn, no counterpart
- Executable check
- 2026 within ±5%, skipped without microdata
- Population costings are survey-weighted estimates; the pipeline produces no interval (limits).
- HMRC’s June 2025 ready reckoner is post-behavioural on administrative SPI data. Ours is static and survey-based, so this is a benchmark, not a validation.
- A static costing below a post-behavioural one runs against intuition. The paper’s leading explanation is the data basis: administrative data capture the top of the income distribution more completely than a survey. Nothing yet decomposes the gap.
- The reckoner stops at 2028–29, so the 2030 figure has no counterpart.
- Every figure comes from hosted runs under policyengine 6.11.1; job ids are in
papers/pe-microsim/figures/basic_rate_costing.json. Under policyengine 4.x they were £6.46bn and £7.38bn. - The executable check is
test_population_impact_uk_real_basic_rateinintegration/tests/test_population.py.
Limits
What the model leaves out, and the data and versions it runs on.
| Limit | Detail |
|---|---|
| Static, no general equilibrium | Prices, wages and output do not move; behavioural responses are optional and post-hoc. |
| UK data vintage | enhanced_frs_2023_24; upstream's FRS 2024-25 vintage is not adopted here. |
| Survey-based aggregates | Population estimates carry sampling, imputation and ageing error, and no interval is computed for them. HMRC's costings use administrative Survey of Personal Incomes data and embed taxable-income elasticities, so a few hundred million of difference on a broad-based rate change is expected. |
| Gated UK microdata | Population runs need a HUGGING_FACE_TOKEN with access. Household calculations are unaffected. |
| Statute-vintage divergence with OLG | This package tracks the latest release; the OLG member is pinned to policyengine-uk==2.88.0, so baseline statute can disagree — a declared inconsistency. |